Computer vision systems it more easily recognize moving items than stationary ones, such as a moving car or a pedestrian in a crosswalk, according to research from Carnegie Mellon University’s Robotics Institute. The Computer vision project was supported by Toyota Research Institute and was worked on by Zhipeng Bao, a robotics PhD student, and Martial Hebert, dean of CMU’s School of Computer Science and a professor in the Robotics Institute. The study may improve how computers and robots recognize things in videos automatically.
Since object identification is essential for comprehending real-world scenarios, creating motion-guided approaches to item discovery could enhance autonomous driving. Additionally, it might be helpful for domestic robots, manipulating robots, and retail robotics. The CMU researchers created a framework called MoTok in collaboration with colleagues from Toyota, the University of California, Berkeley, and the University of Illinois Urbana-Champaign that enables the computer to recognize characteristics of objects it sees moving on its own. MoTok then reconstructs the object using these features, enabling the computer to find the thing in a way that permits it to locate the identical object once more.
Since then, the researchers have expanded their study to enable the computer to virtualize and simplify these aspects. This advancement helps the computer to more accurately recognize high-level traits, allowing it to categorize items rather than only identify a specific object. On the arXiv preprint service, the paper is currently accessible. “Obviously, that doesn’t scale,” Hebert said.
Get the full details of this post in the link below:
